Gallery · LIBERO-Pro

Open the top layer of the drawer and put the bowl inside

libero_goal_object, task 3 · seed 2. DeepSeek-V4.1-Flash, code mode: 10 decisions, 26 motions, solved.

2× · 64 s of motion
  1. 1 of 10 · grasp1.py
  2. 2 of 10 · pull2.py
  3. 3 of 10 · pull3.py
  4. 4 of 10 · open_look.py
  5. 5 of 10 · lift1.py
  6. 6 of 10 · extract2.py
  7. 7 of 10 · bowl_grasp.py
  8. 8 of 10 · descend1.py
  9. 9 of 10 · bowl_grasp2.py
  10. 10 of 10 · place1.py
  1. set_gripper
  2. move_pose
  3. move_pose
  4. set_gripper
  5. move_pose
  6. move_pose
  7. move_pose
  8. move_pose
  9. move_pose
  10. set_gripper
  11. move_pose
  12. move_pose
  13. move_pose
  14. move_pose
  15. move_pose
  16. move_pose
  17. move_pose
  18. move_pose
  19. set_gripper
  20. move_pose
  21. set_gripper
  22. move_pose
  23. move_pose
  24. move_pose
  25. move_pose
  26. move_pose
Decisions

The recording is the robot's own cameras at 2×; the time between motions, while the model thinks, is not filmed. The 3D replay is the camera's picture in depth after each decision, with the paths it planned, to turn and step through.